class: center, middle background-image: url(data:image/png;base64,#https://www.amscins.com/wp-content/uploads/2011/05/Stockholm-university.jpg) background-size: 125px background-position: 5% 92% # Group assignment ### Facial Trustworthiness Predicts Extreme Criminal-Sentencing Outcomes - Wilson & Rule (2015) Alicia Lenander<br>Aleksandra Mandić<br>Eva Dickmänken<br>Hannah Logemann<br>Jón Ingi Hlynsson #### Stockholm University #### Date last edited: 2021-12-26 --- # (a) Causal problem In Wilson's (2018) study 1, the researchers hypothesized that inmates who look less trustworthy were more likely to receive death sentences. The research's aim is to infer the *population average treatment effect* (PATE) of receiving a sentence among potential inmates as predicted by: ??? In this presentation we will go over the main results, highlight its main findings, strengths and weaknesses -- - Facial trustworthiness, - Afrocentricity, - Attractiveness, - Facial maturity, - Facial width-to-height ratio, - Presence of glasses, and - Presence of tattoos. --- ## Boxplot for outliers .pull-left[In the boxplot we can see that there are no substantial outliers in the data.] .pull-right[ <!-- --> ] --- # (b) Main variables relevant to the causal problem .pull-left[**Logistic Regression** **Independent variable**: Trustworthiness (1 = not at all trustworthy, 8 = very trustworthy) **Dependent variable**: Sentence (0 = Life, 1 = Death) ] -- .pull-right[**Covariates**: - Afrocentricity (we assume: 1 = not at all afroc., 8 = very afroc.) - Attractiveness (we assume: 1 = not at all attractive, 8 = very attractive) - Facial Maturity (we assume: 1 = not at all attractive, 8 = very attractive) - Facial width-to-height ratio (fWHR) - Presence of glasses (1=Yes, 0=No) - Presence of tattoos (1=Yes, 0=No) ] --- ### Effects plot of the ANOVA .pull-left[ As can be seen in the figure, there was no interaction between race and trustworthiness. Hence, race was omitted from the **logistic regression**. ] .pull-right[ <!-- --> ] --- class: center top # Study design The study is a a natural (*quasi-experimental*) observational experiment.  --- # (d) Threats to validity and design tricks for improved validity. - No information concerning the scales for afrocentricity, attractiveness, facial maturity - Few participants with tattoos and glasses -- Design element: **Nonequivalent dependent variable** -- - e.g., how ***friendly*** does the inmate look - e.g., how ***happy*** does the inmate look ??? We will talk more about threats to internal validity in the **weeknesses** section. only 14 targets with tattoos and 7 with glasses A nonequivalent dependent variable is a design element suggested by Campbell and is defined as something that's affected by the treatment condition but independent of the dependent variable. HOWEVER, friendliness and trust probably correlate - BUT we would need to test that. Essentially make the raters judge them on perceived happiness and trustworthiness or freindliness and trustworthiness. --- ## Logistic regression models **Table 1. Odds ratios and compatibility intervals** <table class=" lightable-classic-2 table" style="font-family: Times; margin-left: auto; margin-right: auto; font-family: Times; "> <thead> <tr> <th style="text-align:left;"> </th> <th style="text-align:left;"> OR </th> <th style="text-align:center;"> 2.5 % </th> <th style="text-align:center;"> 97.5 % </th> </tr> </thead> <tbody> <tr grouplength="2"><td colspan="4" style="border-bottom: 0;"><strong>Model 1</strong></td></tr> <tr> <td style="text-align:left;padding-left: 2em;" indentlevel="1"> (Intercept) </td> <td style="text-align:left;"> 2.68 </td> <td style="text-align:center;"> 1.30 </td> <td style="text-align:center;"> 5.58 </td> </tr> <tr> <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Trustworthiness </td> <td style="text-align:left;"> 0.70 </td> <td style="text-align:center;"> 0.54 </td> <td style="text-align:center;"> 0.90 </td> </tr> <tr grouplength="8"><td colspan="4" style="border-bottom: 0;"><strong>Model 2</strong></td></tr> <tr> <td style="text-align:left;padding-left: 2em;" indentlevel="1"> (Intercept) </td> <td style="text-align:left;"> 8.83 </td> <td style="text-align:center;"> 1.82 </td> <td style="text-align:center;"> 44.16 </td> </tr> <tr> <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Trustworthiness </td> <td style="text-align:left;"> 0.66 </td> <td style="text-align:center;"> 0.50 </td> <td style="text-align:center;"> 0.89 </td> </tr> <tr> <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Afrocentricity </td> <td style="text-align:left;"> 0.79 </td> <td style="text-align:center;"> 0.67 </td> <td style="text-align:center;"> 0.92 </td> </tr> <tr> <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Attractiveness </td> <td style="text-align:left;"> 0.85 </td> <td style="text-align:center;"> 0.64 </td> <td style="text-align:center;"> 1.12 </td> </tr> <tr> <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Maturity </td> <td style="text-align:left;"> 0.87 </td> <td style="text-align:center;"> 0.73 </td> <td style="text-align:center;"> 1.04 </td> </tr> <tr> <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Facial width-to-height ratio (fWHR) </td> <td style="text-align:left;"> 1.39 </td> <td style="text-align:center;"> 1.18 </td> <td style="text-align:center;"> 1.64 </td> </tr> <tr> <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Presence of glasses </td> <td style="text-align:left;"> 1.56 </td> <td style="text-align:center;"> 1.02 </td> <td style="text-align:center;"> 2.41 </td> </tr> <tr> <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Presence of tattoos </td> <td style="text-align:left;"> 0.58 </td> <td style="text-align:center;"> 0.19 </td> <td style="text-align:center;"> 1.76 </td> </tr> </tbody> </table> --- ## (e) Key result - **Model 1**: Targets perceived as less trustworthy were more likely to be sentenced to death. - **Model 2**: Along with trustworthiness, Afrocentricity was negatively associated with the death sentence and fWHR and the presence of glasses were positively associated with the death sentence. --- ## (f) Effect size estimate(s) **ANOVA** - `\(\eta_{\text{p}}^{2}\)` - Since partial eta squared always carries a positive bias, they could have also reported Omega squared Logistic regression - `\(\Delta{\text{x}^{2}}\)` and Odds ratio. - Odds ratio was used as effect size for the individual predictors tested - CI for Odds ratio was reported - `\(\Delta{\text{x}^{2}}\)` was used to test the fit of the models - `\(R^2\)` could have been reported to show variance elucidation in general ??? OR > 1 means an increasement in Odds, and OR < 1 means a decreasement in Odds `\(\Delta{\text{x}^{2}}\)`: (Does the data deviate significantly from the assumptions of the model? -> Significance shows that the model does not fit well) --- class: inverse center middle # Strengths and weaknesses. --- # Strengths -- **Real inmates**: Construct validity & External validity ??? -> People who have actually been convicted are used in the study. Actual measurement of inmates means that we are measuring people who have been convicted -- **Matching**: Internal/Construct validity ??? -> They have equal groups and pave the way for making causal inferences -- **Grayscale images**: Construct validity ??? -> Death sentence and life sentence inmates wear different colored clothes; thus grayscale images hinder the possibility of participants guessed/knew what sentence inmates received a priori. This constitutes blinding of outcome among raters. -- **Well defined sample**, restricted on sex and race: Internal validity/Construct validity ??? Almost complete "population" of death sentenced inmates in sample (given sex and race). Big sample size (observations among participants) which increases our confidence in the causal conclusion. -- **Different raters for trustworthiness and covariates**: Construct validity ??? The ratings of trust don't affect the ratings of the covariates (continuation of "biased" pattern does not continue (IF there was a pattern)) --- # Weaknesses -- **Unequal inmates with glasses and tattoos** making them hard to infer about ??? All types of validity -> - 14 people with tattoos [7/7 in each group] - 115 with glasses [50/65 in each group] - 742 observations in total -- **Low Cronbach’ s Alpha**: Statistical conclusion validity ??? -> .72 for trust which has the highest coefficient of determination `\(R^2\)` Note: 8 point scale, it is perhaps unintuitive – however using a finer scale leads to a loss in variability in the answers. -- **Info given to participants unclear**: Internal/Construct validity ??? Hard to replicate this study, even given data and OS, on the basis that we have no clue what prior to the study about the inmates -- **Photos taken after sentencing**: Internal validity ??? -> Perhaps a death sentence has an effect on perceived trustworthiness compared to life sentence. -- - *Reverse causation* is a possibility here. -- **Age not controlled for**: Internal/Construct validity ??? Potential confounder; IF it is a confounder, then it is related to BOTH outcome) and exposure --- class: inverse center middle # Thank you for your attention